OK, so I managed to get a bit further by sucking patches from darcs.haskell.org into my tree.  (Incidentally, the tree is a pretty pristine one, obtained by unpacking a tarball.)

But then I got the failure below.  Something bad happens in the nofib directory.  Incidentally its defaultrepo file contains
                http://darcs.haskell.org/nofib
as one would expect.

So darcs crashes again, with the disturbing message that it's messed up the repo.

I tried running darcs repair as the message suggests.  But *that* crashed too.

Let's try darcs check. Nope, that crashes too (in the same way, this time).

It seems that virtually anything I do makes darcs crash.   This is the kind of thing that makes me feel so uneasy about darcs.

All with darcs2 on Windows.

Simon

== running darcs pull -av --repodir nofib
Reading inventory of repository c:/simonpj/darcs/validate-HEAD/nofib inventory
Pulling from "http://darcs.haskell.org/nofib"...
This is the nofib darcs repository (HEAD branch)

For more information, visit the GHC developer wiki at
  http://hackage.haskell.org/trac/ghc
**********************
We have the following new (to them) patches:
They have the following patches to pull:
Tue Jun  3 15:58:57 GMT Daylight Time 2008  Ian Lynagh <igloo@earth.li>
  * Add some x86/Linux outputs
Tue Jun  3 17:49:37 GMT Daylight Time 2008  Ian Lynagh <igloo@earth.li>
  * Add x86-linux output for a couple of benchmarks running the slow way
Getting and merging the following patches:
Tue Jun  3 15:58:57 GMT Daylight Time 2008  Ian Lynagh <igloo@earth.li>
  * Add some x86/Linux outputs
Tue Jun  3 17:49:37 GMT Daylight Time 2008  Ian Lynagh <igloo@earth.li>
  * Add x86-linux output for a couple of benchmarks running the slow way
diffing dir...
Applying to pristine... Tue Jun  3 15:58:57 GMT Daylight Time 2008  Ian Lynagh <igloo@earth.li>
  * Add some x86/Linux outputs

darcs failed:  takeFile ./real/fem/fem.stdout-x86-linux-0 in c:/simonpj/darcs/validate-HEAD/nofib/_darcs/pristine: openFd: does not exist (No such file or directory)
Your repository is now in an inconsistent state.
This must be fixed by running darcs repair.

Tue Jun  3 17:49:37 GMT Daylight Time 2008  Ian Lynagh <igloo@earth.li>
  * Add x86-linux output for a couple of benchmarks running the slow way
darcs failed: 512 at ./darcs-all line 49, <IN> line 44.

sh-2.04$ darcs2 repair
Repairing patch 4/344

darcs failed:  File './spectral/cichelli/Aux.hi' already exists!

sh-2.04$ darcs2 check
Checking patch 5/344 : Mon Jan  8 20:13:28 GMT Standard Time 1996  partain
Unapplicable patch:
Mon Jan  8 20:13:28 GMT Standard Time 1996  partain
  * [project @ 1996-01-08 20:13:28 by partain]
  Initial revision

darcs failed:  File './spectral/cichelli/Aux.hi' already exists!
sh-2.04$

Attachments

• unnamed

The takeFile error sounds like another case of darcs not supporting Windows well
enough.

I'm going to create a new bug report for the darcs check/repair failure.  It
looks like something interesting may be going on there.

On second thought, let's keep this all on one bug for a moment.

I've retrieved  a copy of this repository via darcs get.  Darcs check and repair
report no problems (darcs 2.0.2 + 23 patches).  Also, this does not appear to be
related to case sensitivity in any way.  Something else is going wrong.

One open question is if the takeFile error is related to the darcs check/repair
failure.

I need a volunteer to do the following:
 1. create a tarball of darcs.haskell.org:/home/darcs/nofib
 2. retrieve and unpack said tarball
 3. on Windows: perform darcs check and repair using darcs 2

The purpose here is to see if darcs check and repair work without the takeFile
issue confusing things.

Ok, we still need a volunteer, but for less work.

Ian has created this tarball:
 http://darcs.haskell.org/kowey/nofib.tar.gz

On Linux, it darcs checks fine (provided you remove _darcs/current to avoid a
"Multiple pristines" error).

So it would be helpful if somebody on Windows could just download that file,
unpack it and do a darcs check (removing the redundant _darcs/current)

Ok, Arjan has found the source of the darcs check failure.  Please see the
discussion on http://bugs.darcs.net/issue53 for details.

In short, darcs check/repair just won't work on Windows, but it'll be fine on
any other operating system such as MacOS X (note that case is not the issue
here; see issue53 for details).

I'm leaving this bug open because of the takeFile issue.

OK so this bug was largely about the darcs check failure that Simon got.  I
claim that this is resolved with Darcs-2.3 fetching old-fashioned repositories
as hashed by default (for the GHC team to benefit from this their clean tarballs
will have to be hashed format).

I've split off the takeFile bug into issue1532